Skip to content
All work

Applied AI

Proposal

Private On-Premises AI Platform

On-premises LLM inference, so confidential engineering data can be used with AI without leaving the building.

On sitePubliccloud

512 GB

Unified memory for local models

0

Confidential data sent off site

Role
Research, specification and business case
Stack
Apple SiliconLocal LLMsEXOEntra ID

The problem

Teams wanted the productivity of modern AI tools, but OEM engineering data is confidential and the automotive supply chain expects strict information security. Sending that data to a public service was not an option.

The approach

  • Benchmarked distributed inference on Apple Silicon, comparing a multi-node cluster against a single high-memory machine.
  • Specified a single 512 GB unified-memory node as the simplest, most cost-effective route to large local models.
  • Framed the case around data sovereignty and information security requirements.
  • Designed a self-hosted front end with chat, document templates and presentation tools behind Microsoft sign-in.

The outcome

  • Business case presented to the board and taken forward to security review.
  • A route to AI adoption with no confidential data leaving site.

What they say

Mike stood up a private AI service on hardware in our building, with a chat interface for staff and the model endpoint never exposed. It means people can use AI on our own information without any of it leaving the company. For a business with customer IP in every file, that distinction matters.
Nigel LempriereCTO and Founder, TECNIQ

Want something like this for your business?

Get in touch